Speaker recognition based on subspace methods